Quantifying detrimental risks from exposure to hazardous agents is an important component in the process of risk evaluation and analysis. The focus of this exploratory project is to develop and study new methods of statistical inference for use in quantitative risk assessment. Application is directed to biomedical, occupational, environmental, toxicological, or pharmacological studies where human or animal data are used to set benchmark or other safe, low-dose levels of a hazardous agent, but where study information is limited to high- dose levels of the agent. The resulting guidelines can help improve public health planning and risk regulation when dealing with low-level exposures to hazardous stimuli. Emphasis will be on estimation of the benchmark dose (BMD) associated with a pre-selected level of benchmarked risk (BMR), based on risk/safety data in the form of proportions. Specific focus in this exploratory project centers on ways to avoid the problems of model-dependency and model selection bias, by basing estimation and inferences on model-independent isotonic regression techniques. The theoretical, asymptotic characteristics of the model-free point estimators will be established, and from these large-sample benchmark dose confidence limits (BMDLs) will be constructed. Software developed for BMD estimation and for calculating the confidence limits will be distributed on the Internet, to allow access to the methods by the widest possible corps of users. The methods will fill existing gaps in model-independent methods for benchmark analysis, and will have application to the important problem of modern low-dose risk/safety assessment with proportion data. An evaluation phase of the project will apply the methods to an assortment of risk analytic, quantal response data, assembled from the public domain/existing literature, and will study the small-sample operating characteristics of the methodology via Monte Carlo computer calculations. This will assess the operating capabilities of the new methods and determine if further advances in benchmark analysis can be based on this new estimation paradigm. PUBLIC HEALTH RELEVANCE: This proposal initiates and explores development of model-independent statistical approaches for setting benchmark or other safe low-dose exposure levels of a hazardous stimulus when existing data are limited to high-dose levels of the agent. Common in occupational, environmental, toxicological, and pharmacological risk/safety studies, the methods expand upon traditional benchmark risk analyses and provide insight into two scientifically pressing issues: (i) that most existing benchmark estimation techniques depend heavily on the forms chosen to model the stimulus/dose response;and (ii) that in the face of potential model misspecification/model selection bias, existing statistical strategies for benchmark dose estimation can yield inaccurate, and possibly unsafe, low-dose inferences. This work will explore development of robust benchmark dose inferences, from which resulting low-dose guidelines can improve public health planning and risk regulation when dealing with low-level exposures to hazardous stimuli.